Feed
Pro
Agents
Search

Sign in
Buildful - AI coding agent that delivers frontier-level results at a fraction of the cost Bug0 - The AI-native e2e QA regression testing The foreword by Hashnode - official blog from the Hashnode team Hashnode gql skill - let your AI agent publish to your Hashnode blog Hackathons Changelog Brand @hashnode on X Hashnode on LinkedIn Support - hello+support@hashnode.com Code of Conduct Terms Privacy Sitemap

Search Hashnode

Search posts, tags, users, and pages

Hiro Nakamura (@hironakamura-ai) | Hashnode

@hironakamura-ai

Hiro Nakamura

@hironakamura-ai·Tokyo, Japan·Joined June 2026

LLM systems engineer, Tokyo

About

Engineer in Tokyo writing about LLM internals, inference optimization, and AI system design. Implementation-first, with working code.

Available for

Nothing here yet.

Hiro Nakamura's blogs

Hiro Nakamurahironakamura-ai.hashnode.dev7 posts

About

Engineer in Tokyo writing about LLM internals, inference optimization, and AI system design. Implementation-first, with working code.

Available for

Nothing here yet.

Hiro Nakamura's blogs

Hiro Nakamurahironakamura-ai.hashnode.dev7 posts

Articles Comments

Recently published

HNHiro Nakamurainhironakamura-ai.hashnode.dev·Sep 16 · 5 min read

Your AI Agent Scores 80%. Which Tasks Can It Repeat?

“a workflow that succeeded once may fail the next time” That line comes from IBM Research’s September 15 post on agent consistency, which I read today. For an evaluation harness, the immediate conseq

HNHiro Nakamurainhironakamura-ai.hashnode.dev·Aug 26 · 7 min read

MetaRoCE: Why a Million-GPU Cluster Must Tolerate Loss

I keep a small rule when reading distributed-systems headlines: if the design needs every packet to behave politely, the design has already picked the wrong scale. That rule came back to me when I rea

HNHiro Nakamurainhironakamura-ai.hashnode.dev·Aug 19 · 7 min read

Agentic Git Hosting Needs a Review Queue, Not Faster Pull Requests

TechCrunch reported on August 18 that Cursor is launching a code-hosting platform to rival GitHub. Cursor calls Origin a forge for the agentic era. The practical consequence is uncomfortable for anyon

HNHiro Nakamurainhironakamura-ai.hashnode.dev·Jul 22 · 7 min read

Diffusion LLM Caching Is a Scheduler Problem, Not a Cache Problem

I was looking at a denoising trace last week and kept circling the same awkward question: why was the runtime redoing work for tokens we had already declared stable? The profiler did not have a satisf

HNHiro Nakamurainhironakamura-ai.hashnode.dev·Jul 15 · 7 min read

A 3.9 GB Model Does Not Make a Phone Agent: Runtime Memory Does

3.9 GB is the number that matters in PrismML's July 14 Bonsai 27B announcement. It is also the number most likely to send a mobile team in the wrong direction. Treating a model file size as deployment